Preferred Networks Visual Inspection has a proven track record of being applied to various materials such as metal parts, cloth/fibers, food, resin, rubber, electronic parts, glass, wood, and wafers. We introduce some examples of how this product can find various defects in various materials, with images.
In all cases, we trained with dozens of images of both good and defective products and visualized the results of applying the trained model to separate, unused images as a heatmap.
* The workpieces and defects in each image are virtual ones for the purpose of example and are rendered with computer graphics. Also, some noise is added to each image to create variations. This product was applied to those images for training, detection, and visualization.
Resin (Injection Molding)
This is a case where this product was applied to resin (plastic) products formed by injection molding.
It can find minute weld lines and dirt while ignoring variations in surface texture, lighting conditions, and slight positional shifts. It can be seen that it reacts strongly only to defects and not to shadows or brightness changes in the grooves and edges of the workpiece.
Metal (Forging)
This is a case where this product was applied to the head of a forged screw (with a Phillips-head).
It finds minute cracks while ignoring surface irregularities, the characteristics of the Phillips-head, and light reflections.
Metal (Casting)
This is a case where this product was applied to the machined surface of a cast metal part.
It can be seen that it reacts only to blowholes while ignoring confusing color unevenness, machining marks, and positional shifts on the machined surface.
Cloth / Fiber
This is a case where this product was applied to the surface of a felt fabric.
It accurately finds only stains while distinguishing the various patterns formed by the fibers.